nima.hejazi
← Back to blog

Building ask: An LLM in Your Terminal

Python CLI LLM Terminal

ask is a Python CLI for natural language interaction with LLMs. This post covers why it exists and the choices that keep it fast and out of the way.

Installation

Homebrew (macOS)

brew tap nimahejazi/tap
brew install nh-ask-cli

You may need brew trust nimahejazi/tap.

pip

pip install nh-ask-cli

From Source

git clone https://github.com/nimahejazi/ask.git
cd ask
pip install -e .

Why a terminal assistant?

Most days I have a question that is faster to ask than to grep the docs for: how do I revert one file in git, what port is this process holding, why is this log complaining. The friction was leaving the shell, opening a browser, and pasting back a command. ask collapses that loop to a single line.

The first version did one thing: stream an answer. Two flags made it genuinely useful. -c returns a single bare command for easy copying, and together with the interactive --it mode they turn “I wonder how” into “it is done” without copy paste gymnastics.

Keeping the model local or cloud

A terminal tool should not lock you into one provider. Run ask once to configure your preferred LLM: Mock, Ollama, LM Studio, Anthropic, or ChatGPT. Settings are stored and can be overridden with environment variables.

ask ──► local (Ollama / LM Studio) | cloud (Anthropic / OpenAI) | custom

That split means the same CLI works on a laptop with no API key and on a machine wired to a cloud model.

Tool support

The -t ./tools.sh flag lets you pass shell functions as tools. The model calls them when appropriate, extending its capabilities without hardcoding logic.

Architecture in brief

The code is plain Python. ask/ holds the CLI entrypoint and provider clients; the chat loop handles streaming responses and tool orchestration. Configuration is resolved from environment variables, and installation supports Homebrew, pip, or editable installs from source.

stdin / args ──► ask ──► provider ──► streamed answer or command
                        └──► tool execution (when requested via -t)

Tests run with pytest or python -m pytest, and the package builds with pip install ..

##Wrapping up

ask is deliberately small. It answers, it returns a command you can run, and it extends itself via tools when needed. The result is a tool that stays in the background until you actually need it.